Embed Size (px)
Citation preview
Detection of the Adulteration of Fresh Coconut Water via NMR Spectroscopy and Chemometrics
Paul I. C. Richardson1, Howbeer Muhamadali1,2, Yang Lei1, Alexander P. Golovanov1*, David I. Ellis1*, Royston Goodacre1,2*
1School of Chemistry, Manchester Institute of Biotechnology, 131 Princess Street, University of Manchester, UK, M1 7DN.
2Department of Biochemistry, Institute of Integrative Biology, University of Liverpool, Biosciences Building, Crown Street, Liverpool, L69 7ZB, UK
Email: [email protected], [email protected] , [email protected]
Keywords: coconut water, food fraud, adulteration, sugars, dilution, masking, NMR, chemometrics
1
1
2
3
45
67
89
10
1112
13
1415
16
Abstract
Here, we applied NMR spectroscopy in combination with chemometrics to quantify the
adulteration of fresh coconut water, stretched with water-sugar mixture. Coconut water was
extracted from young Costa Rican coconuts and adulterated with concentrations of various
sugar solutions. A total of 45 samples were analysed by 1D proton NMR spectroscopy and
chemometrics. Results showed highly sensitive quantification, with a limit of detection of
adulteration with sugars of 1.3% and a root-mean-squared error of prediction of 0.58%.
Interestingly, we observed a regular drift in the chemical shift and change in the lineshape of
malic acid signals concomitant with increasing levels of adulteration was identified. On
further investigation this was found to originate from changes in the concentration of divalent
cations such as magnesium within the samples. It can be concluded that NMR enables
accurate quantification for the degree of adulteration in this product, with the added discovery
finding that the shift and lineshape of malic acid signal can be utilised as a potential
diagnostic marker for partial substituting of fresh coconut water with extrinsic components
such as sugar mixtures.
2
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
1.Introduction
Coconut water is a refreshing beverage obtained by extracting the liquid endosperm from
immature (6-9 months) coconut fruits (cocos nucifera). While all parts of the fruit and tree
have been utilised throughout history within their native tropical regions for a multitude of
purposes including building, horticultural substrate1, or emergency ersatz intravenous drip2,3,
the high mineral content in coconut water along with its suggested health benefits have led to
the drink being advertised as a natural alternative to isotonic sports drinks. These factors have
contributed to a rapid expansion in the coconut water market, with sales in the UK reaching
over £100 M in 20164; a 20-fold increase since 2012.
The sudden popularity of this product can lead to the risk of demand outstripping supply, as
coconut palms require several years of maturation before they are able to bear fruit5.
Furthermore, with the vast majority of coconut water coming from only five countries
worldwide (Brazil, Sri Lanka, Indonesia, Philippines and Thailand), these factors can result in
coconut water being vulnerable to food fraud6,7.
The characterisation of coconut water has been undertaken using several different methods.
High-performance liquid chromatography (HPLC) and liquid chromatography mass
spectrometry (LC-MS) have been used to analyse multiple classes of phytohormones8, and
headspace solid phase micro-extraction gas chromatography (HS-SPME-GC) used to
characterise the volatile organic profile. Nuclear magnetic resonance (NMR) spectroscopy
combined with chemometric methods has also been applied to analyse and monitor the effect
of different industrial processing methods9. However, the novelty of coconut water and recent
popularity has meant that the detection of fraud in this product has not as yet been studied in
detail, with only one study published to date10.
3
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
Moreover, fraudulent activity involving coconut water products has already been observed11–
13. With an investigation in late 2017 by the National Food Crime Unit of the UK Food
Standards Agency (FSA) showing that at least 14 major coconut water products contained
undeclared sugars14. These were detected using isotopic ratio mass spectrometry (IRMS)15
and site-specific natural isotope fractionation NMR (SNIF-NMR)16. While these are powerful
and versatile method for the absolute detection of adulteration with sugars17, they require
lengthy preparation times (24 & 72 hours respectively), and are thus poorly suited as
screening methods.
Here, we used 1H NMR and chemometric methods to detect product dilution masked with a
normalised sugar profile substitution. Fresh coconut water was adulterated at various
concentrations using a buffered sugar solution emulating the natural concentrations of
glucose, fructose, and sucrose present in our sample. In order to achieve our aim for a
screening method, minimal sample processing was performed. Furthermore, we investigated
a systematic drift observed for the signals belonging to malic acid with increasing levels of
substitution with sugar solution, which revealed its origin in the changing metal coordination
state of malate caused by such an adulteration. This effect on malic acid signal shift and
lineshape has the potential to be used as a diagnostic marker for tampering with coconut
water products.
2. Methodology
Preparation of coconut water:
Seven immature coconuts (6-9 months) originating from Costa Rica were obtained from a
UK-based online retailer. The juice was extracted from each of these using a specialised
4
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
Cocodrill® device obtained from the same retailer, centrifuged at 18,000 g, for 10 min at 4°C
and pooled together to make a consistent stock solution. The pooled solution was then stored
at -80°C in 45 mL aliquots until required. Prior to use, the coconut water was thawed at room
temperature and separated into 1 mL aliquots, which were heat-treated using a TechneDri-
Block DB-3A (Cole-Parmer, Staffordshire, UK) hot-plate for 150 s at 70 °C to simulate an
industrial pasteurisation process, recombined and cooled in a refrigerator at 5 °C. Heat-
treated aliquots were used for a maximum of three days after being thawed. Sugar solutions
were made using D-glucose anhydrous, D-sucrose (analytical grade for biochemistry 99%,
RNAse and DNAse free), and D-(-)-fructose 99%≤ purchased from Fisher Scientific (Fisher
Scientific Ltd. Loughborough, Leics, UK), Acros Organics (Acros Organics, Geel, Belgium)
and Sigma-Aldrich (Sigma-Aldrich Chemie GMBH, Stanheim, Germany) respectively.
K2HPO4 and KH2PO4 were obtained from Alfa Aesar (Alfa Aesar, Haysham, United
Kingdom).
Adulteration of coconut water
Fructose (12.44 g), glucose (13.29 g) and sucrose (5.77 g) were dissolved in deionised water
(500 mL) to make a solution emulating the natural concentrations of each of these sugars in
coconut water (24.88, 26.58 and 11.54 mg.mL-1 respectively) obtained in our previous
study10. To ensure that the pH and conductivity remained constant at all levels of adulteration,
1 M K2PO4 (0.177 mL) and 1 M KH2PO4 (4.724 mL) were added to 45.1 mL sugar solution
(pH: 5.31, conductivity: 5.69 mS.m-1). Aliquots of heat-treated coconut water were then
adulterated with this buffered solution (0-100% adulteration, in 5% increments) each with a
total volume of 2 mL. Each sample was spiked with 50 μL of a 4,4-dimethyl-4-silapentane-1-
sulfonic acid (DSS) (Sigma-Aldrich) in D2O solution (Sigma-Aldrich) (98.17 mg.mL-1) to
provide chemical shift reference as well as a signal for radiofrequency lock. Finally, prior to
analysis, samples were divided into three 0.6 mL aliquots. Excluding one sample at 75%
5
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
adulteration that was rendered unusable due to human error, all prepared samples were
analysed using 1H NMR spectroscopy, leading to a total of 62 spectra for this experiment.
NMR analysis:
Proton 1D spectra were acquired at 20°C using a Bruker 800 MHz NMR AVANCE III
spectrometer (Bruker, Coventry, UK) equipped with a 5mm TCI cryoprobe with temperature
control unit using standard pulse program “zgesp” from the Bruker library. A relaxation
delay of 1.4 s and an acquisition time of 1.47 s were used with 16 scans, and exponential
window function with 1 Hz line broadening was applied prior to Fourier transformation in
Topspin 3.5 (Bruker). All NMR spectra were referenced to the DSS peak at 0.00 ppm.
Addition of Mg2+ ions to a sodium malate solution:
A 11.88 mg.mL-1 stock solution of malic acid (Sigma-Aldrich) was made, as well as a 0.5 M
solution of MgCl2 (Fischer Scientific). Buffered malate solution meant to approximate that
found in coconut water was made by combining 2896 µL of the malic acid stock solution,
824 µL 1M KH2PO4 solution, 26 µL 1M K2HPO4 solution, 563 µL 1 M KOH (Fischer
Scientific) solution and 5711 µL H2O (3.40 mg.mL-1 malate18, pH: 5.32). The stock solution
was then spiked with 25 µL DSS solution. An 800 µL aliquot of malate solution was added to
an NMR tube and analysed using NMR as a control sample, then charged with 5 µL of the
MgCl2 solution. It was analysed again, then sequentially charged with 10 µL aliquots of
MgCl2 solution five times (end volume of MgCl2 added: 55 µL). This final solution was
diluted in a malate-free buffer solution (stock: 824 µL 1M KH2PO4 solution, 26 µL 1M
K2HPO4 solution, 9150 µL H2O) to 75, 50, 25, 12.5, and 6.25% concentration(s), and each of
these was analysed using 1H NMR spectroscopy. A total of 12 spectra were obtained.
Data analysis:
6
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
Fourier transformed NMR spectra were imported to Matlab and truncated to include only the
spectral region between 5.6004 and -0.0503 ppm containing the main peaks (these comprised
13277 data points per spectrum). Baseline correction was then performed using a de-trending
algorithm, spectra were-scaled using the peak intensity of internal reference signal of DSS,
and the spectral region containing water signal around 4.8 ppm removed. When required, the
processed spectra were further divided into two spectral sections at 3.2 ppm. Principal
components analysis (PCA) was employed to reduce the dimensionality within the spectral
datasets into uncorrelated principal components (PCs) and describe relationships between
samples. A linear predictive model, used for quantification, was also obtained using partial
least squares regression (PLSR)19. The spectral dataset was divided into a training-set
containing spectra for adulteration levels starting at 0% in 10% increments of increasing
concentration (0%, 10%, 20%, 30%, up to 100%) and a test-set containing spectra for
adulteration levels starting at 5% in 10% increments of increasing concentration (e.g. 5%,
15%, 25%, 35%, up to 95%). Several predictive models were created using the training-set,
with each of the models using an increasing number of latent variables (LVs). During
calibration each model (on the training set only) was then tested using a k-fold cross-
validation (k=20) algorithm, and the optimal model was chosen by aiming to minimise the
number of LVs along with the root-mean-squared error of cross-validation (RMSECV). Once
chosen, the predictive model was evaluated by challenging the model with the tests-set.
To investigate the relationship between the chemical shift of malic acid signals and
adulteration, maxima locations for the strongest peaks in the signal at 2.7 ppm were found
using the Matlab “findpeaks” command. These data were then imported into Microsoft Excel
and a least-squares linear regression algorithm was used to acquire a line of best fit. Due to
the lack of malic acid in the 100% adulteration samples, these spectra were not included in
these calculations.
7
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
3. Results and Discussion
Although the full spectral range measured during experiments was 12 to -3 ppm, the vast
majority of strong signals were contained within 5.4 to 0 ppm. To minimise the effect of
noise and simplify further calculations, spectra were therefore truncated to an effective
spectral range of 5.6 to -0.05 ppm. Furthermore, the high concentrations of sugar in coconut
water relative to its other constituents led to these signals being orders of magnitude more
intense and subsequently overpowering statistical models. To remedy this and obtain a clear
understanding of the effect of variance within other coconut water constituents, spectra were
further divided into two subsections: the first containing the signals associated with sugars
(5.6 to 3.2 ppm) and the second containing signals associated with other coconut water
constituents (3.2 to -0.05 ppm). Both of these sections were then analysed with PCA (Figure
1).
PCA scores plot for the coconut water constituents region (Figure 1A) showed very little
variance within adulteration levels– indicating that replicate measurements were highly
reproducible. While PC 1 shows a clear and constant trend from positive to negative scores
with increasing adulteration, PC 2 shows a "boomerang effect", with the scores increasing
until approximately 50% adulteration and a gradual decrease to return to their initial negative
scores as the level of adulteration increases.
Loadings in PC 1 (Figure S1B) closely resemble the NMR spectra (Figure S1A), indicating
that the trend is an overall decrease in intensity, with the only peaks not present being those
ascribed to the internal DSS standard at 0.00, 0.63, 1.75, 2.91, and 3.14 ppm. This trend is
expected, as virtually all the signals in this region are ascribed to coconut water, and
increasing the level of adulteration will consequently decrease the intensity of those peaks.
8
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
The loadings of PC 2 (Figure S1C) however, show the main cause for this boomerang effect
to be due to the two signals at ~2.70 and ~2.45 ppm assigned to malic acid 20. Along with a
decrease in intensity, a decrease in chemical shift occurs with increasing levels of
adulteration as well as signal sharpening (Figure 4). For both signals, this is reflected in the
loadings by negative weighting at the more de-shielded part of the NMR spectra in samples
with lower adulteration levels, and a positive weighting with the effect of decreasing intensity
of the signals.
PCA scores plot for the sugar signal section (Figure 1B) also shows a clear trend related to
adulteration, this time along a composite axis combining PC 1 and PC 2, allowing for a clear
separation between different levels of adulteration. Unlike the coconut water constituents
section, however, the sugar signals section shows significant variance within classes,
indicating that, predictably, the detection is not as obvious. Nonetheless, the ability to detect
even small discrepancies readily in the sugar profile using NMR is promising. Interestingly, a
further examination of the loadings plot allows for the discrimination of signals for each
sugar and provides further details concerning the imperfections in our masking of dilution.
Figure 2 provides an enhanced view of the signals associated with glucose, fructose and
sucrose (4.3-3.2 ppm). Although the overlaid spectra (Figure 2A) show a high signal density
in this area, the loadings plot of PC 1 (Figure 2B) combined with the NMR spectrum of each
sugar allows for the assignment of peaks and thus a description of the imperfections in
masking. For example, our mixed solution was found to contain slightly less glucose and
fructose, while containing more sucrose. Moreover, utilising this information could
potentially allow for the differentiation between natural variation in the sugar profiles of
coconut water samples and the intentional addition of sugars.
Quantitative analysis
9
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
In addition to exploratory analysis, a quantitative predictive model was created using partial
least-squares regression (PLSR). PLSR was performed on both sugar signals and coconut
water constituent subsections, along with the full effective spectral range. Plots of each model
are presented in Figure 3, while a summary of the overall statistics of PLS in terms of
linearity, error and detection limits is presented as Table 1.
It is immediately clear upon observing the resulting plots (Figure 3) that the combination of
NMR and PLSR allow for the creation of a powerful predictive model. Both individual
subsections, along with the full effective range, showed high accuracy within both the test
and cross-validation sets even at low levels of adulteration and despite the naturally occurring
sugar profile being emulated within the adulterant.
The strength of this analytical model is further confirmed by the various statistical results
(Table 1). Based on the R² and Q² values for the cross-validation and test sets consistently
reaching >0.99, these data demonstrated a very good fit to the linear predictive models.
Additionally, the RMSE on the test sets are all within 1%, indicating that the models can
quantitatively predict the level of adulteration within 5%. Two models are presented for the
coconut water region, one using 3 LVs and one using 5 LVs. While the optimal model can
generally be discerned from the plot comparing the RMSECV to the number of latent
variables used in each model (data not shown), both could be viewed as acceptable without
the requirement for supplementary testing. It is also interesting to compare the prediction
error of the whole spectral range to that of the individual regions; while one would intuitively
expect to see either an average of the two or, similarly to the PCA scores, a peak intensity-
based weighting leading to equivalent results to the sugar section, the combined sections
seem to create a more powerful model than its individual components (with the provision that
the 3 LV model is correct for the coconut water section). A possible reason for this may be a
higher number of peaks that can be used to validate the model’s predictions, thus lowering
10
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
random uncertainty. While increasing the number of data points used in the spectra also often
increases the contribution of background noise, the signal-to-noise ratio in these spectra is so
high that this drawback is not significant. Furthermore, the limits of detection (LODs), which
were calculated using a comparison of the predicted and real values on the test set with a
false positive/negative rate of 0.05, were found to be within 2%, reaching 0.6% for the 5LV
coconut constituent region model. Given the results of each model, we recommend the use of
the entire dataset (5.6-0 ppm) rather than individual parts to ensure that all factors relevant to
adulteration are taken into account.
Previously, we presented Raman spectroscopy for the first time as an effective screening
method for the detection of adulteration in coconut water, as it is an inexpensive, portable and
very fast method, requiring virtually no sample preparation10. Using this vibrational
spectroscopic method, we showed it was possible to detect adulteration by dilution and its
masking with single sugars down to 1.9%. However, due to its reliance on discrepancies in
the sugar profile to adulteration, the ability of Raman spectroscopy to detect fraudulent
dilution of coconut water was hampered by normalising the individual concentrations of
glucose, fructose, and sucrose throughout the adulteration. The use of NMR spectroscopy,
though more time and resource intensive, would detect virtually all cases of intentional
substitution while retaining the advantage of minimal sample preparation, as lower levels of
adulteration would likely not be commercially worth the risk inherent to the practice.
Malate signal drift analysis
While the peaks present in the NMR spectra generally presented changes in intensity related
to adulteration, three signals (~2.4, ~2.7, and ~4.3 ppm) in the spectra presented an additional
drift in chemical shift and change in lineshape with respect to the addition of increasing
levels of sugar solution (Figure 4). All of these signals were attributed to malic acid or
11
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
malate, an organic di-carboxylic acid (and its conjugate base) naturally present in coconut
water. Each malate signal shows a similar unique increase in chemical shift and signal
broadening with increasing concentration of coconut water that is not observed for any other
signals. When plotted against the proportion of coconut water in the samples, the chemical
shifts of two visible peaks in the 2.7 ppm range depicted a linear drift (Figure 4D & E).
Three hypotheses were postulated and tested to explain the potential source(s) of this
behaviour: (i) slight variations in pH; (ii) the variation of concentration of malic acid itself;
and (iii) variation of divalent cation concentrations leading to differences in malate metal
coordination state. The pH hypothesis was tested by altering the pH of coconut water
aliquots using 1M solutions of NaOH and HCl (Figure S3), while the concentration
hypothesis was tested by diluting a buffered 3.4 mg.mL-1 solution of sodium malate,
emulating the natural concentrations in coconut water (Figure S4). Neither of these led to any
significant drift in the malate peaks and were therefore ruled out. Further details on these
experiments can be found in the Supplementary Information.
To test the divalent cation coordination cation hypothesis, separate aliquots of buffered
sodium malate solution were spiked with increasing concentrations of zinc (in the form of
ZnCl2) and magnesium (in the form of MgCl2), two metal cations naturally present in coconut
water, and the drift in chemical shift of the malate signal at ~2.7 ppm examined. All analysis
was performed on this signal as it was the most sharply defined and contained a consistent
double doublet; however, a similar drift was present in all three malate signals.
The presence of divalent cations led to a significant increase in chemical shift. The addition
of MgCl2 solution to the sodium malate solution led to the malate signal to shift to more
positive values, which is in agreement with the effect observed with increasing
concentrations of coconut water. Furthermore, based on the linear interpolation of the fit, the
12
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
amplitude of the drift fits with that found in coconut water. The concentrations of the two
main divalent cation constituents in coconut water, calcium and magnesium, add up to a total
of ~17 mM: similar addition of MgCl2 would lead to an approximate shift of 0.017 ppm,
based on the results presented in Figure 5B, which matches the total magnitude of 0.017-
0.022 ppm drift observed as a result of adulteration (Figure S3). The agreement between the
shifts in adulteration experiments and separate malate titration with divalent metal ions
provides strong evidence to defend our hypothesis that the signal shift is due to changed
metal coordination state of malate. Furthermore, increasing the concentration of Zn2+ (Figure
S5) to 10 mM led to a drift of 0.05 ppm in the chemical shift; although the concentration of
zinc cations in coconut water is <0.1 mM, meaning the absolute shift as a result of the
presence of Zn cations is likely not very significant, the presence of this effect provides
further evidence for these cations being the source of the drift.
A final confirmation of our hypothesis was acquired through the dilution of the final MgCl2-
spiked sodium malate solution with buffer solution (Figure 6). Although the change in
chemical shift was less linear than that observed in the spiking experiments, the chemical
shift did return close to its original position prior to spiking with divalent cations (Figure 5A).
Interestingly, while the source of the signal drift has now been accounted for, the relative
signal broadening of malate peaks when the higher percentage of coconut water is present in
the sample is currently unclear but can likely be explained by further malate binding to other
yet unknown natural ingredients of coconut water. Importantly, malate signals serve as a
sensitive reporter for the relative concentration of this genuine product in adulterated
samples.
4. Conclusion
13
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
Coconut water is a refreshing and nutritious beverage which has gained huge popularity in
recent years4. This sudden increase in popularity and sales, however, has greatly increased the
vulnerability and potential risk of illicit behaviours such as economically motivated
adulteration (food fraud). Further, this relatively recent and sudden increase in popularity and
sales of this product could be said to result in there not currently being a sufficient range and
application of testing methods available to food regulatory bodies, which instead rely on
IRMS and SNIF-NMR. Here, we proposed a combination of NMR and chemometrics to
create a powerful model which, importantly, requires virtually no sample preparation. Using
this approach, we were able to detect substitution of coconut water with a sugar solution
emulating the natural concentrations of glucose, fructose and sucrose (this drink’s major
soluble solids) at levels as low as 1.3%, while being able to quantify it to within 0.6% error.
Additionally, we investigated a regular linear drift observed in the chemical shift of malate
peaks as the percentage of adulteration increased. While changes in pH and concentration of
malate were not found to be significant, changes in the concentration of 2+ cations was found
to cause a drift in our malate solution that is in complete agreement with our findings in
coconut water spectra. Chemical shift being an absolute measurement, this property could be
utilised as a diagnostic marker21 of adulteration indicative of the intentional substitution, or
stretching, of coconut water with foreign, cheaper liquids.
Acknowledgements:
DIE and RG thank the UK ESRC and UK FSA for funding (Food fraud: a supply network
integrated systems analysis (Grant number ES/ M003183/1)).
14
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
References:
1 S. Ikram, U. Habib and N. Khalid, Pakistan J. Agric. Sci., 2012, 49, 121–125.
2 G. A. Petroianu, M. Kosanovic, I. Saad Shehatta, B. Mahgoub, A. Saleh and W. H. Maleck, J. Trace Elem. Exp. Med., 2004, 17, 273–282.
3 B. Eiseman, AMA Arch Surg., 1954, 68, 167–178.
4 J. Glotz, Infographic: The UK coconut water market at a glance, https://www.thegrocer.co.uk/buying-and-supplying/categories/soft-drinks/infographic-the-uk-coconut-water-market-at-a-glance/541126.article, (accessed 12 October 2017).
5 P. Batugal and R. Bourdeix, in Coconut Genetic Resources, eds. P. Batugal, V. R. Rao and J. Oliver, 2005, pp. 251–267.
6 D. I. Ellis, H. Muhamadali, S. A. Haughey, C. T. Elliott and R. Goodacre, Anal. Methods, 2015, 7, 9401–9414.
7 D. I. Ellis, V. L. Brewster, W. B. Dunn, J. W. Allwood, A. P. Golovanov and R. Goodacre, Chem. Soc. Rev., 2012, 41, 5706–5727.
8 Z. Ma, L. Ge, A. S. Y. Lee, J. W. H. Yong, S. N. Tan and E. S. Ong, Anal. Chim. Acta, 2008, 610, 274–281.
9 N. R. Sucupira, E. G. Alves Filho, L. M. A. Silva, E. S. de Brito, N. J. Wurlitzer and P. H. M. Sousa, Food Chem., 2017, 216, 217–224.
10 P. I. C. Richardson, H. Muhamadali, D. I. Ellis and R. Goodacre, Food Chem., , DOI:10.1016/j.foodchem.2018.08.038.
11 L. Sorias, It’s now Solo ‘coconut flavoured’ water, http://www.trinidadexpress.com/20160617/news/its-now-solo-coconut-flavoured-water, (accessed 18 September 2017).
12 Food Standards Agency, V Fresh Coconut Milk Drink withdrawn because it contains milk, https://www.food.gov.uk/news-updates/news/2015/14760/v-fresh-coconut-milk-drink-withdrawn-due-to-undeclared-milk, (accessed 18 September 2017).
13 Food Standards Agency, Interlink Direct recalls its Orthodox Coconut Palm brand Coconut Juice due to undeclared milk, https://www.food.gov.uk/news-updates/news/2015/14774/interlink-direct-recalls-its-orthodox-coconut-palm-brand-coconut-juice-due-to-undeclared-milk, (accessed 18 September 2017).
14 J. Glotz, FSA probe finds widespread addition of undeclared sugar in coconut water, https://www.thegrocer.co.uk/buying-and-supplying/categories/soft-drinks/fsa-probe-finds-widespread-addition-of-undeclared-sugar-in-coconut-water/558787.article.
15 E. Jamin, R. Guérin, M. Rétif, M. Lees and G. J. Martin, J. Agric. Food Chem., 2003, 51, 5202–5206.
16 I. J. Košir, M. Kocjančič, N. Ogrinc and J. Kidrič, Anal. Chim. Acta, 2001, 429, 195–
15
321
322
323324
325
326327328
329330
331332
333334
335336
337338
339340
341342343
344345346
347348349350351
352353354
355356
357
206.
17 G. G. Martin, Y. L. Martin, N. Naulet and H. J. D. McManus, J. Agric. Food Chem., 1996, 44, 3206–3213.
18 U. Santoso, K. Kubo, T. Ota, T. Tadokoro and A. Maekawa, Food Chem., 1996, 57, 299–304.
19 P. S. Gromski, H. Muhamadali, D. I. Ellis, Y. Xu, E. Correa, M. L. Turner and R. Goodacre, Anal. Chim. Acta, 2015, 879, 10–23.
20 G. Campo and J. I. Santos, 2006, 556, 462–468.
21 D. I. Ellis, H. Muhamadali, D. P. Allen, C. T. Elliott and R. Goodacre, Curr. Opin. Food Sci., 2016, 10, 7–15.
16
358
359360
361362
363364
365
366367
368
369
Figures and tables:
Figure 1: PCA scores plots for adulteration of coconut water with our mixed sugar solution.
To obtain a meaningful representation of the variance within different regions of the spectral
range, spectra were truncated into two sections prior to analysis: a sugar region (5.00 – 3.20
ppm) containing higher intensity peaks originating from glucose, sucrose and fructose; and a
coconut region (3.20 –-0.05 ppm) containing signals originating from other components of
coconut water. A: PCA scores plot for the coconut region, obtained with two PCs and 89.7%
total explained variance (EV). B: PCA scores plot for the sugar region, obtained with two
PCs and 88.4% TEV. The inset colour bars represent the percentage level of adulteration.
17
370
371
372
373
374
375
376
377
378
379
380
Figure 2A : 1H NMR spectra for the adulteration of coconut water with our mixed sugar
solution ranging from 0% (blue) to 100% (red) adulteration. Spectra have been truncated to
show only the main saccharide signals (4.25-3.20 ppm); the inset colour bars represent the
percentage level of adulteration. B: PC 1 loadings plot for the spectral dataset above,
demonstrating the variance associated with each chemical shift measured. Negative loadings
indicate a higher intensity at that chemical shift at lower adulteration levels, while positive
loadings indicate a higher intensity at higher adulteration levels. NMR spectra for sucrose
(blue), glucose (green) and fructose (red) obtained from literature are also presented in B.
18
381
382
383
384
385
386
387
388
389
390
391
392
Table 1: PLSR results for the optimal models of the sugar region, coconut region, and the
combined spectral range. LVs represents the number of latent variables used in each model,
R² and Q² represent the goodness of the model’s fit to the dataset, while trn, CV and tst
denote results pertaining to the training, cross-validation and test set, respectively. The limit
of detection (LOD) is also presented. For the coconut (CW) region, the results of two models
with different number of LVs used in model construction are presented as both seemed
equally viable.
Method LVs
R² Q² (CV)
Q²(tst)
RMSE (trn, %)
RMSE (CV, %)
RMSE (tst, %)
LOD (%)
Sugar region 3 0.9992 0.9991 0.9991 0.8863 0.7810 0.8416 1.9836
CW region 3 0.9994 0.9989 0.9991 0.7631 0.8426 0.8515 1.6790
CW region 5 0.9999 0.9995 0.9999 0.2993 0.5541 0.2974 0.6513
Full Range 4 0.9998 0.9995 0.9996 0.4948 0.5889 0.5767 1.3418
Figure 3: PLSR results plots comparing the predicted actual concentrations for the coconut
region (A), sugar region (B) and combined spectral range (C). For Figure 3A, the model with
3 LVs was used.
19
393
394
395
396
397
398
399
400
401
402
403
404
405
Figure 4A-C: 1H NMR spectra for the adulteration of coconut water with our mixed sugar
solution ranging from 0% (blue) to 100% (red) adulteration, zoomed in on each peak
assigned to malate. 4D-E: chemical shift of malic acid peaks plotted against level of
adulteration (0% being pure coconut water, 100% being pure sugar solution) for the malate
signal at ~2.7 ppm, along with regression results.
20
406
407
408
409
410
411
412
413
Figure 5A: Overlaid 1H NMR spectra depicting the effect on the ~2.7 ppm malate
signalmultiplet of the incremental addition of aqueous Mg2+ (MgCl2) to a buffered 3.44
mg.mL-1 solution of sodium malate ranging from 0 mM Mg2+ (blue) to 31.43 mM Mg2+ (red).
B: Plot comparing the drift in chemical shiftof each peak in the malate signal
multipletrelative to its corresponded initial position prior to addition of Mg2+ ions.
21
414
415
416
417
418
419
420
Figure 6A: Overlaid 1H NMR spectra depicting the effect of dilution of the sample containing
sodium malate mixture with MgCl2. The malate signal at ~2.7 ppm was monitored, with
spectra overlaid for various proportions of these original components, from 100% in the
undiluted sample to 6.25% for the most diluted one. B: Plot comparing the drift in chemical
shift of each peak in the malate signal multipletrelative to its position in the undiluted sample.
22
421
422
423
424
425
426
